Watson and the Case for AI Staffing

21

 Graphic from Raymond Kurzweil

This article builds on some of my earlier posts which dived into the likely prominence of artificial intelligence in supply chain management (and visibility) in the future. Specifically, we’ll look at the 2010 performance of the IBM question-answering AI system called “Watson” and discuss what it portents for supply chain careers.

Introducing “Watson”

Last year, while visiting some friends in the USA, I got a suggestion to look at the Fall 2010 issue of AI Magazine, which had an article titled “Building Watson: An Overview of the DeepQA Project”. The article covered an as-yet quiet project at IBM to create an artificial intelligence system which could compete in Jeopardy. For those who live outside North America, Jeopardy is a long-standing quiz show on television. The contestants listen to a statement (formed as an answer), have the option to press a buzzer, and then the first player to “buzz in” can provide the appropriate question matching to the original statement. Here is an example statement: “The current president of the united states”, which would be answered by the person who presses their buzzer first as “Who is Barak Obama”. If a player responds incorrectly, they lose money. Being a good player at Jeopardy requires three things:

  1. Broad knowledge, retrieved by vague references in many formats (pictures, sound, quotes, puns, jokes, etc)
  2. High capability to judge the probable correctness of your answers (to avoid penalties for incorrect responses)
  3. Being fast at the two skills above

Since the article published in the Fall of 2010, the IBM system (named “Watson”) went on to compete in 55 matches against human champions. Not only did the system win, but it won with tremendous margins. In the final matches against the standing champions of Jeopardy, Watson earned 66% of the entire points for the game while the other two players earned 16% each. In 1997 another IBM machine of related design, named DeepBlue, settled the long standing challenge of para-human AI performance in chess. It bested Garry Kasparov, the world champion at the time. Now the same has occurred with a much more ambiguous (and frankly useful) task.

Is Watson a Surprise?

In many ways, Watson fulfills predictions about how AI will come into our lives in the “para-human” stage. In this stage, AI is expected to have similar overall levels of intelligence as humans, but with some individual cognitive aspects being better and others being worse when directly compared to human capabilities. Consider, for example, that Watson’s “learning” encompasses tremendous volumes in short periods of time. But it can only learn from very structured inputs. Another example is the sharp contrast between its abilities in information management vs. creativity. Watson knows every single word in Wikipedia, but cannot draw inferences from it to create new knowledge. Reportedly, Watson can process 500 gigabytes, the equivalent of a million books, per second. A great learner by design, Watson nonetheless gives us nothing back but our own insights, hidden from humans by a need for overly tedious searching or a shortage of time in which to find a usable answer.

There are other ways in which Watson aligns with AI expectations: the one I am most interested in is the net cost of Watson compared to its capabilities. A capable, well-educated supply chain analyst could expect to earn about 65,000 USD per year in the USA in 2010. Assuming taxes and raises in-line with inflation, we could say that their 5 year net cost to the business is ~.4 million USD. IBM has stated that building a similar system as Watson would cost 1.5-2 million USD.  This suggests to me that Watson should be replacing at least 4-5 full time, well qualified analysts in our industry in order to be cost neutral over 5 years (and that is frankly the longest anyone should be planning when considering tech investments). In order to be attractive as an investment, it probably needs to replace twice as many staff, so around 8-10 qualified analysts for 5 years. Given its capabilities and limitations, this seems about right. An AI solution like Watson will not run a whole company department, or render most positions redundant. But it can replace many analysts, including senior analysts with very good analytical skills. And given its speed and scalability, it’s more likely to replace a multiple of individuals from the same role than to be replacing one person from many different roles.

AI and Immediate Staff Impacts

For most companies, I don’t believe an AI application will be replacing staff in 2011. But, from a pure capabilities viewpoint it should be clear that staffing changes can already take place. Many, many supply chain analyst jobs are effectively covered by researching existing data, compiling as needed, and presenting results. This includes many pricing analysts, transport analysts, warehouse analysts, etc. Professional services firms like KPMG, Accenture, Price Waterhouse Coopers, etc, must employ hundreds or even thousands of such positions globally. It’s important to note that these analyst positions are fairly high value, even if they have limited “creativity” involved. For professional services firms, this work is billable for 200 to 300 USD per hour. An introduced AI system would be generating respectable revenue.

 

Watson and Future Staff Impacts

The success of Watson portents the changes we should see in supply chain management departments in the next 15 years. Technologically, para-human AI systems of considerable power will be available to most businesses soon, and the pressure to adopt them will be fierce. Part of the competitive pressure to adopt AI into supply chain management will come from AI applications which strike directly at financially important tasks. As an example, the DeepBlue project created an expensive but reliable way of winning chess matches, even against the absolute best competitors in the world. The Watson system shows that a sufficiently interested company can purchase the best question-answering performance from an AI system. Question-answering is an important task in some industries (like clinical medicine, paralegal, fraud detection, etc). Question-answering is of medium value in supply chain management, not low-value but also not what leads to success or failure of a whole company. Now, imagine an application which could negotiate contract pricing better than any human. The AI software might read facial signals, body posture, voice cues, research market pricing instantly, bluff with more force, or calculate the outcome probabilities with more precision.  However it would work, such an application would be (a) astoundingly valuable perhaps even to the point of causing bankruptcies to the losers, and (b) would cause immediate pressure on competitors to adopt AI.

In particular, there is a lesson we should have already learned from technology progression: the arrival of Watson re-affirms that human’s will directly compete with AI in the supply chain field for jobs in the future. Consider it this way:

  • Five hundred years ago, good advice to young people (in Europe anyway) would be to avoid being a scribe because the moveable type press would overcome their role
  • Around two hundred years ago, good advice to young people would have been to avoid cottage industry because it was in a losing fight with industrial mass production.
  • One hundred years ago, going into equestrian industries would have been a mistake
  • Fifty years ago, typing was about to become the slowest way to reproduce text
  • Twenty-five years ago, learning mainframe computing was a bad idea for a career

These examples show places where the technology existed and was ramping up to directly compete for work roles, but people didn’t always see what it meant for their career path. In 2011, I believe, young professionals in supply chain management office roles should be careful to develop themselves past “knowledge worker” into one of the three areas which should retain a “human advantage” longer than others: (a) the ability to learn a novel process very quickly and from ambiguous inputs, (b) creatively producing new knowledge, or (c) roles that require human ethical or legal responsibility.

Monday Morning Wrap Up:

As with the other articles on this site, I am closing with a summary of how you can use this information in your daily working life. Today’s topic continued on previous discussions of how artificial intelligence will become a major concern in supply chain management (including supply chain visibility, probably before other areas). Here are the points of immediate use:

  • Artificial intelligence is coming to supply chain management (in some places, it’s already here in 2011).
  • Humans will compete with AI for some roles, just as we competed with any number of industrial machines and computer applications (Xerox machines, word processors, even pocket calculators).
  • A new, public business-case has occurred with the Watson win in Jeopardy which you can use to open discussions with people unfamiliar with, or doubtful of, this trend
  • The recent case of IBM’s Watson is not so much a surprise as an affirmation of what we expected from the first wave of AI. Namely, they will be:
    • In total, similar to human capabilities, but vastly over or under humans in specific attributes
    • Focused on knowledge management, integration, extraction, etc. But NOT creating new knowledge
    • Not adaptable to ambiguous, novel processes but rather to standard processes with defined inputs
    • AI purchasing costs are comparable to replacing a small human team. As system capabilities increase, the average salary level of the team (and the size of the team being replaced) are likely to increase.
    • Given that humans will compete with AI for some roles, it is worthwhile to ensure your career track gets you off of the competitive overlap space as quickly as possible. Even if you get out, you should realize how devalued those roles will appear later in time. For example, think about what “typist” or “mainframe administrator” looks like on a job candidate background today.

Leave a Reply